TRefine, Control and Distill: A Text-to-Image Framework for Faithful Image Generation